Carnegie Mellon Univ. v. Marvell Tech. Grp., Ltd.

Decision Date23 September 2013
Docket NumberCivil Action No. 09–290.
Citation986 F.Supp.2d 574
PartiesCARNEGIE MELLON UNIVERSITY, Plaintiff, v. MARVELL TECHNOLOGY GROUP, LTD. et al., Defendants.
CourtU.S. District Court — Western District of Pennsylvania

OPINION TEXT STARTS HERE

David T. McDonald, Douglas B. Greenswag, Nicola J. Templeton, Theodore J. Angelis, K & L Gates LLP, Seattle, WA, Eliza K. Hall, Patrick J. McElhinny, Christopher M. Verdini, Joseph J. Porcello, Mark G. Knedeisen, Roberto Capriotti, K & L Gates, Pittsburgh, PA, for Plaintiff.

David C. Radulescu, Gregory S. Maskel, Joseph Milowic, Kathleen M. Sullivan, Quinn Emanuel Urquhart & Sullivan, New York, NY, John E. Hall, Timothy P. Ryan, Eckert, Seamans, Cherin & Mellott, Pittsburgh, PA, Andrew J. Bramhall, Quinn Emanuel Urquhart & Sullivan LLP, Redwood City, CA, Brian E. Mack, Melissa J. Baily, Quinn Emanuel Urquhart & Sullivan, LLP, San Francisco, CA, Derek Shaffer, Quinn Emanuel Urquhart & Sullivan LLP, Washington, DC, Edward J. Defranco, Faith E. Gay, Raymond N. Nimrod, Robert Wilson, Quinn Emanuel Urquhart & Sullivan, LLP, New York, NY, Pro Hac, Vice, Heather E. Belville, Kevin P. Johnson, Mark Tung, Melissa Chan O'Sullivan, Quinn Emanuel Urquhart & Sullivan LLP, Redwood Shores, CA, Steven G. Madison, Quinn Emanuel Urquhart & Sullivan, LLP, Los Angeles, CA, for Defendants.

OPINION

NORA BARRY FISCHER, District Judge.

I. INTRODUCTION

This is a patent infringement case brought by Plaintiff, Carnegie Mellon University (CMU), against Defendants Marvell Technology Group, Ltd. and Marvell Semiconductor, Inc. (collectively Marvell), alleging that Marvell has infringed two of its patents, U.S. Patent Nos. 6,201,839 (the “'839 Patent”) and 6,438,180 (the “'180 Patent”) (collectively, the “CMU Patents”). CMU contends that Marvell's infringement was willful. (Docket No. 461). Marvell counters that the CMU Patents are invalid. (Docket No. 465). This matter was tried before a jury for four weeks, with jury selection starting on November 26, 2012. (Docket No. 760). A number of motions for Judgment as a Matter of Law (“JMOL”) were made before the verdict was rendered. (Docket Nos. 699; 701; 703; 731; 738; 740; 742; 747). The Court denied these motions on the record 1 on December 21, 2012. (Docket No. 759). The case was then presented to the jury. After deliberations, the jury entered a verdict on December 26, 2012 in favor of CMU on infringement, validity, and willfulness, awarding damages in the amount of $1,169,140,271.00 (Docket No. 762).

Following the trial, the Court entertained post-trial motions, wherein the parties: (1) renewed their earlier JMOL contentions; (2) moved for a new trial on several grounds; (3) argued the equitable defense of laches; and (4) requested a permanent injunction, post-judgment royalties, supplemental damages, interest, enhanced damages, as well as attorney fees.2 (Docket Nos. 786–811). These matters have been completely briefed (Docket Nos. 823–829; 832–837; 849–855; 857–863), and the Court heard argument on same from May 1 through May 2, 2013. (Docket No. 873). 3 The Court writes now to explain its reasoning for denying the pre-verdict motions for JMOL, and to rule on the renewed JMOLs, the Motions for New Trial, and Motion for a Remittitur.

II. FACTUAL BACKGROUND4
A. Technology in Suit

The patents-in-suit are generally directed to the method of sequence detection in high density magnetic recording sequence detectors. See '839 Patent col. 16 ll. 20–23.

1. Hard Disk Drive Data Recordings

Hard disk drives (“HDD”) contain a platter or disk that holds data on concentric tracks. (Docket No. 673 at 154). The device bears a visual resemblance to the classic record player. ( Id.). Just as a record player has a needle attached to the tip of the arm, an HDD has a “read head” that reads and writes data onto these tracks. ( Id.). Each track is made up of a track width, and this track width is broken into millions of bit regions. ( Id.). The track is made of magnetic material. ( Id.). The bit regions are magnetized to store data in the form of “zeros” and “ones.” ( Id.). As the track moves underneath the read head, the read head picks up the fields emanated from these magnetic regions on the track and turns the fields into read back signal samples. ( Id. at 155). However, the read back signal samples are not exactly equal to what is actually written on the disk. ( Id. at 156). For instance, the read back signal may read “0.3” when a “zero” was written on the track. ( Id.). These discrepancies occurring during the read back process are referred to as “noise.” ( Id.).

2. Viterbi–Like Detector and the Trellis Concept

A Viterbi-like read channel detector found in the HDD takes the read back signal samples and determines the sequence of symbols written on the disk using a trellis. ( Id. at 157–158). This process is called “sequence detection.” ( Id. at 158). A trellis section is used to represent a string of bits sitting on a medium. ( Id.). There are four potential sequences of two bits, called states: 01, 11, 00, 10; and they can be connected by branches. ( Id. at 162–163).

A trellis is used to represent a string of these bits; for example, a three-bit string of 011, would be represented by a “01” connected by a branch to “11.” ( Id.). One trellis section includes all possible bit sequences. ( Id.). In this instance, a single trellis section of 011, 010, 111, 110, 001, 000, 101, 100 is represented as follows:

Tabular or graphic material set at this point is not displayable.

(Docket No. 771 at Ex. C at 14). A trellis can then be created to represent a sequence of any length. For example, a six bit sequence is represented as follows:

Tabular or graphic material set at this point is not displayable.

( Id. at 18). Through this trellis, one can trace a path that is equivalent to a specific sequence of symbols. ( Id.). For example 100101, is shown below:

Tabular or graphic material set at this point is not displayable.

( Id. at 17).

The detector determines the “best path” through the trellis, meaning the best or most likely written sequence on the disk, using branch metric values. (Docket No. 673 at 169). The read back signal samples are taken by the detector to compute the branch metric.5 ( Id. at 170). The path with the lowest branch metric values becomes the detected sequence. ( Id.). Thus, the detector calculates the path with the lowest cumulative branch metric value to determine the detected sequence of zeros and ones written on the disk. ( Id. at 172).

3. Noise

Bit regions are not homogeneous. ( Id. at 175). Rather, they are made up of small tiles or magnetic grains that create regions of magnetization that do not fall within straight bit regions on the track. ( Id.). As the bit regions become narrower in high density recording and more bits are packed onto a smaller area, there will be fewer grains per bit region. ( Id. at 176). With fewer grains, islands of grains may develop in which the detector cannot accurately read the data. ( Id. at 176–177). This is shown below in a diagram in which green represents “zeros,” and blue represents “ones.” 6

Tabular or graphic material set at this point is not displayable.

(Docket No. 771 at Ex. C at 24). So, as the density of the recording increases, the amount of noise or uncertainty in the signal also increases. (Docket No. 673 at 179). As seen below, the amount of noise is also affected by the specific sequence of bits written on the track.

Tabular or graphic material set at this point is not displayable.

(Docket No. 771 at Ex. C at 25). This is correlated signal-dependent noise, because the noise signals from one boundary to the other move together, either attracting or moving away from each other. (Docket No. 673 at 179).

Noise was previously assumed to be white, or flat, at all time instances and in all branches.7 ( Id. at 183–184). Using this noise assumption in determining disk signals worked in the low density environment of the 1970s and 1980s. ( Id. at 184). A Viterbi-like detector computed Euclidean branch metric values based on the assumption that the noise was white. ( Id.). Next, the industry used another assumption, that of correlated noise, where the noise had structure but the structure was the same regardless of the symbol sequence (i.e., written symbols). ( Id. at 186). The current assumption is that of correlated signal-dependent noise. ( Id. at 193). This is media noise in the read back signal, whose noise structure is attributable to a specific sequence of symbols. ( Id.). Below is a comparison of the three forms:

Tabular or graphic material set at this point is not displayable.

(Docket No. 771 at Ex. C).

4. The CMU Patents

With the last model of signal-dependent noise, the detected sequence is obtained by maximizing the likelihood function. (Docket No. 673 at 206–207). The CMU Patents start by showing that such a likelihood function is dependent on all the read back signals and all written symbols from the entire disk. (Docket No. 673 at 206–207). This is expressed as:

Tabular or graphic material set at this point is not displayable.

'839 Patent Eq. 1.

As there are billions of symbols on the disk, the likelihood function is broken up into smaller per sample functions. (Docket No. 673 at 208). The CMU Patents derived a function based on the observed signal samples; postulated a sequence of written symbols; then applied certain mathematical manipulations to turn the function into a quotient of a likelihood function, as seen below. ( Id. at 214–215).

Tabular or graphic material set at this point is not displayable.

'839 Patent Eq. 4–6. The resulting function can be used to create different embodiments, as disclosed in the CMU Patents. (Docket No. 673 at 220). One embodiment is called the correlation matrices embodiment, expressed in Equation 13 of the '839 Patent:

Tabular or graphic material set at this point is not displayable.

'839 Patent Eq. 13; (Docket No. 673 at 221).

Another form of embodiment is the Finite Impulse Response (“FIR”) embod...

To continue reading

Request your trial
9 cases
  • Carnegie Mellon Univ. v. Marvell Tech. Grp., Ltd.
    • United States
    • United States Courts of Appeals. United States Court of Appeals for the Federal Circuit
    • August 4, 2015
    ...subjectively reckless manner with respect to the risk of infringing the subject patents." Carnegie Mellon Univ. v. Marvell Tech. Grp., 986 F.Supp.2d 574, 633 (W.D.Pa.2013). We reject Marvell's contention that those findings cannot stand.Marvell concedes that its engineers "evaluated Dr. Kav......
  • Opticurrent, LLC v. Power Integrations, Inc.
    • United States
    • U.S. District Court — Northern District of California
    • June 5, 2019
    ...company, was alleged to have infringed two patents relating to semiconductor chips. Carnegie Mellon Univ. v. Marvell Tech. Grp., Ltd., 986 F. Supp. 2d 574, 582 (W.D. Pa. 2013). Marvell manufactured the infringing chips in Taiwan and sold them to makers of hard-disk drives, who incorporated ......
  • Bernard v. BNY Mellon
    • United States
    • U.S. District Court — Western District of Pennsylvania
    • February 7, 2022
    ...... *3-5; see also Berckeley Inv. Group, Ltd. v. Colkitt , 455 F.3d 195, 217 (3d Cir. ... See Carnegie. Mellon Univ. v. Marvell Technology Group, ......
  • Voss ex rel. Situated ex rel. Marvell Tech. Grp., Ltd. v. Sutardja, Case No.: 14-CV-01581-LHK
    • United States
    • U.S. District Court — Northern District of California
    • January 26, 2015
    ...of Pennsylvania accusing Marvell of willfully infringing CMU patents. Compl. ¶ 6; see Carnegie Mellon Univ. v. Marvell Tech. Grp., Ltd. ("CMU I"), 986 F. Supp. 2d 574, 599 (W.D. Pa. 2013) (noting that "CMU filed its complaint in this case on March 6, 2009"). After a four-week trial, the jur......
  • Request a trial to view additional results

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT